Merge branch 'main' into please_dont_modify_this_branch_unless_you_are_just_merging_with_main__

Author: NicolasHug

packaging/post_build_script.sh

@@ -1,5 +1,11 @@
 #!/bin/bash
-LD_LIBRARY_PATH="/usr/local/lib:$CUDA_HOME/lib64:$LD_LIBRARY_PATH" python packaging/wheel/relocate.py
+set -euxo pipefail
+
+if [ -n "${CUDA_HOME:-}" ]; then
+    LD_LIBRARY_PATH="/usr/local/lib:${CUDA_HOME}/lib64:${LD_LIBRARY_PATH}"
+fi
+
+python packaging/wheel/relocate.py
 
 if [[ "$(uname)" == "Linux" && "$(uname -m)" != "aarch64" ]]; then
     extra_decoders_channel="--pre --index-url https://download.pytorch.org/whl/nightly/cpu"

packaging/pre_build_script.sh

@@ -36,7 +36,7 @@ else
   conda install libwebp -y
   conda install libjpeg-turbo -c pytorch
   yum install -y freetype gnutls
-  pip install auditwheel
+  pip install "auditwheel<6.3.0"
 fi
 
 pip install numpy pyyaml future ninja

packaging/wheel/relocate.py

@@ -15,7 +15,10 @@
 
 # Third party imports
 if sys.platform == "linux":
-    from auditwheel.lddtree import lddtree
+    try:
+        from auditwheel.lddtree import lddtree
+    except ImportError:
+        from auditwheel import lddtree
 
 
 ALLOWLIST = {

torchvision/models/optical_flow/raft.py

@@ -486,7 +486,7 @@ def forward(self, image1, image2, num_flow_updates: int = 12):
         batch_size, _, h, w = image1.shape
         if (h, w) != image2.shape[-2:]:
             raise ValueError(f"input images should have the same shape, instead got ({h}, {w}) != {image2.shape[-2:]}")
-        if not (h % 8 == 0) and (w % 8 == 0):
+        if not ((h % 8 == 0) and (w % 8 == 0)):
             raise ValueError(f"input image H and W should be divisible by 8, instead got {h} (h) and {w} (w)")
 
         fmaps = self.feature_encoder(torch.cat([image1, image2], dim=0))

torchvision/ops/boxes.py

@@ -78,7 +78,8 @@ def batched_nms(
         _log_api_usage_once(batched_nms)
     # Benchmarks that drove the following thresholds are at
     # https://github.com/pytorch/vision/issues/1311#issuecomment-781329339
-    if boxes.numel() > (4000 if boxes.device.type == "cpu" else 20000) and not torchvision._is_tracing():
+    # and https://github.com/pytorch/vision/pull/8925
+    if boxes.numel() > (4000 if boxes.device.type == "cpu" else 100_000) and not torchvision._is_tracing():
         return _batched_nms_vanilla(boxes, scores, idxs, iou_threshold)
     else:
         return _batched_nms_coordinate_trick(boxes, scores, idxs, iou_threshold)

torchvision/transforms/v2/_color.py

@@ -134,7 +134,7 @@ def _check_input(
             raise TypeError(f"{name}={value} should be a single number or a sequence with length 2.")
 
         if not bound[0] <= value[0] <= value[1] <= bound[1]:
-            raise ValueError(f"{name} values should be between {bound}, but got {value}.")
+            raise ValueError(f"{name} values should be between {bound} and increasing, but got {value}.")
 
         return None if value[0] == value[1] == center else (float(value[0]), float(value[1]))
 

torchvision/transforms/v2/_geometry.py

@@ -567,7 +567,7 @@ class RandomRotation(Transform):
     Args:
         degrees (sequence or number): Range of degrees to select from.
             If degrees is a number instead of sequence like (min, max), the range of degrees
-            will be (-degrees, +degrees).
+            will be [-degrees, +degrees].
         interpolation (InterpolationMode, optional): Desired interpolation enum defined by
             :class:`torchvision.transforms.InterpolationMode`. Default is ``InterpolationMode.NEAREST``.
             If input is Tensor, only ``InterpolationMode.NEAREST``, ``InterpolationMode.BILINEAR`` are supported.